voicelisted

# Voice — speech in, speech out Two directions, both via ElevenLabs, both with a documented fallback so a missing API key degrades rather than breaks: - **Speech → text**: `scripts/transcribe.sh` — ElevenLabs Scribe first, local `mlx-whisper` (via `uv`, no manual model download) if no key is set. **Apple Silicon only** for the local fallback; without a key on any other platform this fails with a clear message instead of guessing. - **Text → speech**: `scripts/tts-file.sh` — ElevenLabs multilingual TTS. There is no local fallback for this direction (no bundled TTS engine ships with this plugin) — without a key, it fails loudly and tells you where to get one. `scripts/say-voice.sh` wraps this and sends the result straight to Telegram as a voice note (`sendVoice`), for use with the [Telegram bridge](../../bridge/README.md). ## Setup Add `ELEVENLABS_API_KEY` via the `/jarvis-setup` page, or set it directly in `<hub>/.env`: ``` ELEVENLABS_API_KEY=sk_... ``` Get a key at <https://elevenlabs.io/app/settings/api-keys> (free tier available). **Without a key:** `transcribe.sh` still works on Apple Silicon Macs (falls back to local `mlx-whisper`, needs `uv` and `ffmpeg` on `PATH`); `tts-file.sh`/`say-voice.sh` have nothing to fall back to and will say so plainly rather than silently doing nothing. ## Usage ```bash "${CLAUDE_PLUGIN_ROOT}/skills/voice/scripts/transcribe.sh" recording.ogg # auto-detect language "${CLAUDE_PLUGIN_ROOT}/skills/voice/scripts/tr